Python: Bug: Inconsistent behavior of function calling from agent class

[NirajC-Microsoft]

Describe the bug
I'm experiencing an issue where calling a new function from an agent class returns the output of a previously invoked function, leading to inconsistent agent responses.

Expected behavior
When a specific function is called from an agent class, the response should accurately reflect the output of that function, not a previously invoked function. Each function call should produce a fresh, contextually relevant response based on the current invocation

Screenshots
Correct Scenario:
Here assign_mentor gets called and giving correct response

Bug:
Here assign_mentor gets called but giving incorrect response, that response is from previous function called from same agent class.

Platform

• Language: Python
• Source: pip 25.0.1
• AI model: OpenAI:GPT-4o
• IDE: [e.g. Visual Studio, VS Code]
• OS: Windows

Additional context
Add any other context about the problem here.

[Roopan-Microsoft]

Hi @markwallace-microsoft - Any update on this?

[moonbox3]

@NirajC-Microsoft, this looks to be an issue related to specifically using an Azure AI Agent inside of the playground? If so, we'll need to file the issue in the Azure SDK for Python repo. SK provides abstractions for the AzureAIAgent (think of it as a wrapper) but we don't have any control over the actual service calls.

[moonbox3]

As a note: the AzureAIAgent may use previous function results from its thread/history/context if it thinks it can answer it without having to invoke the function again. You may need to experiment with certain instructions to the agent to make sure it invokes the function again each time, and tell it to not re-use existing function results in the context.

[moonbox3]

@Roopan-Microsoft, we haven't received any response from you. If you need further help on this, please either re-open the issue, or file a new one. Thank you.